(122a) Machine Learning at Scale for the Advanced Process Control Toolkit

The last decade has seen an explosion of interest and development in the fields of Big Data, Big Data Analytics, and Machine Learning. Recent research has recognized that classical approaches to realizing Artificial Intelligence (AI) has its limitations, and has instead focused on learning systems based on statistical techniques. This approach has led to the rapidly expanding field of Machine Learning.

Machine Learning relies heavily on numerical optimization; this is similar to the significance of optimization algorithms in advanced process control (APC)/model predictive control (MPC) and Online Optimization. As we know, MPC relies on a Linear program (LP) and/or Quadratic program (QP) while Online Optimization typically utilizes nonlinear optimization algorithms. The fields of APC/MPC and Machine Learning have more in common than is obvious at first glance. In APC/MPC, the optimization objective is to minimize a sum of stage costs (finite or infinite); supervised machine learning, for instance, seeks to minimize a sum of stage empirical risk functions/costs.

The objective of this talk is to demonstrate how large-scale machine learning algorithms can be a useful tool in the APC practitioner’s toolkit. Specific large-scale machine learning algorithms are presented with case studies. The first framework is a learning algorithm that allows easy tuning of the objective function in APC. The second framework focuses on how AI algorithms can be used to build and adapt inferentials and/or property estimation correlations. Finally, the real time asset monitoring problem is discussed. Performance guarantees of the algorithms will be briefly discussed.